GNU Linux-libre 6.8.7-gnu
[releases.git] / fs / xfs / libxfs / xfs_sb.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4  * All Rights Reserved.
5  */
6 #include "xfs.h"
7 #include "xfs_fs.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_bit.h"
13 #include "xfs_sb.h"
14 #include "xfs_mount.h"
15 #include "xfs_ialloc.h"
16 #include "xfs_alloc.h"
17 #include "xfs_error.h"
18 #include "xfs_trans.h"
19 #include "xfs_buf_item.h"
20 #include "xfs_bmap_btree.h"
21 #include "xfs_alloc_btree.h"
22 #include "xfs_log.h"
23 #include "xfs_rmap_btree.h"
24 #include "xfs_refcount_btree.h"
25 #include "xfs_da_format.h"
26 #include "xfs_health.h"
27 #include "xfs_ag.h"
28 #include "xfs_rtbitmap.h"
29
30 /*
31  * Physical superblock buffer manipulations. Shared with libxfs in userspace.
32  */
33
34 /*
35  * Check that all the V4 feature bits that the V5 filesystem format requires are
36  * correctly set.
37  */
38 static bool
39 xfs_sb_validate_v5_features(
40         struct xfs_sb   *sbp)
41 {
42         /* We must not have any unknown V4 feature bits set */
43         if (sbp->sb_versionnum & ~XFS_SB_VERSION_OKBITS)
44                 return false;
45
46         /*
47          * The CRC bit is considered an invalid V4 flag, so we have to add it
48          * manually to the OKBITS mask.
49          */
50         if (sbp->sb_features2 & ~(XFS_SB_VERSION2_OKBITS |
51                                   XFS_SB_VERSION2_CRCBIT))
52                 return false;
53
54         /* Now check all the required V4 feature flags are set. */
55
56 #define V5_VERS_FLAGS   (XFS_SB_VERSION_NLINKBIT        | \
57                         XFS_SB_VERSION_ALIGNBIT         | \
58                         XFS_SB_VERSION_LOGV2BIT         | \
59                         XFS_SB_VERSION_EXTFLGBIT        | \
60                         XFS_SB_VERSION_DIRV2BIT         | \
61                         XFS_SB_VERSION_MOREBITSBIT)
62
63 #define V5_FEAT_FLAGS   (XFS_SB_VERSION2_LAZYSBCOUNTBIT | \
64                         XFS_SB_VERSION2_ATTR2BIT        | \
65                         XFS_SB_VERSION2_PROJID32BIT     | \
66                         XFS_SB_VERSION2_CRCBIT)
67
68         if ((sbp->sb_versionnum & V5_VERS_FLAGS) != V5_VERS_FLAGS)
69                 return false;
70         if ((sbp->sb_features2 & V5_FEAT_FLAGS) != V5_FEAT_FLAGS)
71                 return false;
72         return true;
73 }
74
75 /*
76  * We current support XFS v5 formats with known features and v4 superblocks with
77  * at least V2 directories.
78  */
79 bool
80 xfs_sb_good_version(
81         struct xfs_sb   *sbp)
82 {
83         /*
84          * All v5 filesystems are supported, but we must check that all the
85          * required v4 feature flags are enabled correctly as the code checks
86          * those flags and not for v5 support.
87          */
88         if (xfs_sb_is_v5(sbp))
89                 return xfs_sb_validate_v5_features(sbp);
90
91         /* versions prior to v4 are not supported */
92         if (XFS_SB_VERSION_NUM(sbp) != XFS_SB_VERSION_4)
93                 return false;
94
95         /* We must not have any unknown v4 feature bits set */
96         if ((sbp->sb_versionnum & ~XFS_SB_VERSION_OKBITS) ||
97             ((sbp->sb_versionnum & XFS_SB_VERSION_MOREBITSBIT) &&
98              (sbp->sb_features2 & ~XFS_SB_VERSION2_OKBITS)))
99                 return false;
100
101         /* V4 filesystems need v2 directories and unwritten extents */
102         if (!(sbp->sb_versionnum & XFS_SB_VERSION_DIRV2BIT))
103                 return false;
104         if (!(sbp->sb_versionnum & XFS_SB_VERSION_EXTFLGBIT))
105                 return false;
106
107         /* It's a supported v4 filesystem */
108         return true;
109 }
110
111 uint64_t
112 xfs_sb_version_to_features(
113         struct xfs_sb   *sbp)
114 {
115         uint64_t        features = 0;
116
117         /* optional V4 features */
118         if (sbp->sb_rblocks > 0)
119                 features |= XFS_FEAT_REALTIME;
120         if (sbp->sb_versionnum & XFS_SB_VERSION_NLINKBIT)
121                 features |= XFS_FEAT_NLINK;
122         if (sbp->sb_versionnum & XFS_SB_VERSION_ATTRBIT)
123                 features |= XFS_FEAT_ATTR;
124         if (sbp->sb_versionnum & XFS_SB_VERSION_QUOTABIT)
125                 features |= XFS_FEAT_QUOTA;
126         if (sbp->sb_versionnum & XFS_SB_VERSION_ALIGNBIT)
127                 features |= XFS_FEAT_ALIGN;
128         if (sbp->sb_versionnum & XFS_SB_VERSION_LOGV2BIT)
129                 features |= XFS_FEAT_LOGV2;
130         if (sbp->sb_versionnum & XFS_SB_VERSION_DALIGNBIT)
131                 features |= XFS_FEAT_DALIGN;
132         if (sbp->sb_versionnum & XFS_SB_VERSION_EXTFLGBIT)
133                 features |= XFS_FEAT_EXTFLG;
134         if (sbp->sb_versionnum & XFS_SB_VERSION_SECTORBIT)
135                 features |= XFS_FEAT_SECTOR;
136         if (sbp->sb_versionnum & XFS_SB_VERSION_BORGBIT)
137                 features |= XFS_FEAT_ASCIICI;
138         if (sbp->sb_versionnum & XFS_SB_VERSION_MOREBITSBIT) {
139                 if (sbp->sb_features2 & XFS_SB_VERSION2_LAZYSBCOUNTBIT)
140                         features |= XFS_FEAT_LAZYSBCOUNT;
141                 if (sbp->sb_features2 & XFS_SB_VERSION2_ATTR2BIT)
142                         features |= XFS_FEAT_ATTR2;
143                 if (sbp->sb_features2 & XFS_SB_VERSION2_PROJID32BIT)
144                         features |= XFS_FEAT_PROJID32;
145                 if (sbp->sb_features2 & XFS_SB_VERSION2_FTYPE)
146                         features |= XFS_FEAT_FTYPE;
147         }
148
149         if (!xfs_sb_is_v5(sbp))
150                 return features;
151
152         /* Always on V5 features */
153         features |= XFS_FEAT_ALIGN | XFS_FEAT_LOGV2 | XFS_FEAT_EXTFLG |
154                     XFS_FEAT_LAZYSBCOUNT | XFS_FEAT_ATTR2 | XFS_FEAT_PROJID32 |
155                     XFS_FEAT_V3INODES | XFS_FEAT_CRC | XFS_FEAT_PQUOTINO;
156
157         /* Optional V5 features */
158         if (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_FINOBT)
159                 features |= XFS_FEAT_FINOBT;
160         if (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_RMAPBT)
161                 features |= XFS_FEAT_RMAPBT;
162         if (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_REFLINK)
163                 features |= XFS_FEAT_REFLINK;
164         if (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_INOBTCNT)
165                 features |= XFS_FEAT_INOBTCNT;
166         if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_FTYPE)
167                 features |= XFS_FEAT_FTYPE;
168         if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_SPINODES)
169                 features |= XFS_FEAT_SPINODES;
170         if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_META_UUID)
171                 features |= XFS_FEAT_META_UUID;
172         if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_BIGTIME)
173                 features |= XFS_FEAT_BIGTIME;
174         if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_NEEDSREPAIR)
175                 features |= XFS_FEAT_NEEDSREPAIR;
176         if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_NREXT64)
177                 features |= XFS_FEAT_NREXT64;
178
179         return features;
180 }
181
182 /* Check all the superblock fields we care about when reading one in. */
183 STATIC int
184 xfs_validate_sb_read(
185         struct xfs_mount        *mp,
186         struct xfs_sb           *sbp)
187 {
188         if (!xfs_sb_is_v5(sbp))
189                 return 0;
190
191         /*
192          * Version 5 superblock feature mask validation. Reject combinations
193          * the kernel cannot support up front before checking anything else.
194          */
195         if (xfs_sb_has_compat_feature(sbp, XFS_SB_FEAT_COMPAT_UNKNOWN)) {
196                 xfs_warn(mp,
197 "Superblock has unknown compatible features (0x%x) enabled.",
198                         (sbp->sb_features_compat & XFS_SB_FEAT_COMPAT_UNKNOWN));
199                 xfs_warn(mp,
200 "Using a more recent kernel is recommended.");
201         }
202
203         if (xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
204                 xfs_alert(mp,
205 "Superblock has unknown read-only compatible features (0x%x) enabled.",
206                         (sbp->sb_features_ro_compat &
207                                         XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
208                 if (!xfs_is_readonly(mp)) {
209                         xfs_warn(mp,
210 "Attempted to mount read-only compatible filesystem read-write.");
211                         xfs_warn(mp,
212 "Filesystem can only be safely mounted read only.");
213
214                         return -EINVAL;
215                 }
216         }
217         if (xfs_sb_has_incompat_feature(sbp, XFS_SB_FEAT_INCOMPAT_UNKNOWN)) {
218                 xfs_warn(mp,
219 "Superblock has unknown incompatible features (0x%x) enabled.",
220                         (sbp->sb_features_incompat &
221                                         XFS_SB_FEAT_INCOMPAT_UNKNOWN));
222                 xfs_warn(mp,
223 "Filesystem cannot be safely mounted by this kernel.");
224                 return -EINVAL;
225         }
226
227         return 0;
228 }
229
230 /* Check all the superblock fields we care about when writing one out. */
231 STATIC int
232 xfs_validate_sb_write(
233         struct xfs_mount        *mp,
234         struct xfs_buf          *bp,
235         struct xfs_sb           *sbp)
236 {
237         /*
238          * Carry out additional sb summary counter sanity checks when we write
239          * the superblock.  We skip this in the read validator because there
240          * could be newer superblocks in the log and if the values are garbage
241          * even after replay we'll recalculate them at the end of log mount.
242          *
243          * mkfs has traditionally written zeroed counters to inprogress and
244          * secondary superblocks, so allow this usage to continue because
245          * we never read counters from such superblocks.
246          */
247         if (xfs_buf_daddr(bp) == XFS_SB_DADDR && !sbp->sb_inprogress &&
248             (sbp->sb_fdblocks > sbp->sb_dblocks ||
249              !xfs_verify_icount(mp, sbp->sb_icount) ||
250              sbp->sb_ifree > sbp->sb_icount)) {
251                 xfs_warn(mp, "SB summary counter sanity check failed");
252                 return -EFSCORRUPTED;
253         }
254
255         if (!xfs_sb_is_v5(sbp))
256                 return 0;
257
258         /*
259          * Version 5 superblock feature mask validation. Reject combinations
260          * the kernel cannot support since we checked for unsupported bits in
261          * the read verifier, which means that memory is corrupt.
262          */
263         if (xfs_sb_has_compat_feature(sbp, XFS_SB_FEAT_COMPAT_UNKNOWN)) {
264                 xfs_warn(mp,
265 "Corruption detected in superblock compatible features (0x%x)!",
266                         (sbp->sb_features_compat & XFS_SB_FEAT_COMPAT_UNKNOWN));
267                 return -EFSCORRUPTED;
268         }
269
270         if (!xfs_is_readonly(mp) &&
271             xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
272                 xfs_alert(mp,
273 "Corruption detected in superblock read-only compatible features (0x%x)!",
274                         (sbp->sb_features_ro_compat &
275                                         XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
276                 return -EFSCORRUPTED;
277         }
278         if (xfs_sb_has_incompat_feature(sbp, XFS_SB_FEAT_INCOMPAT_UNKNOWN)) {
279                 xfs_warn(mp,
280 "Corruption detected in superblock incompatible features (0x%x)!",
281                         (sbp->sb_features_incompat &
282                                         XFS_SB_FEAT_INCOMPAT_UNKNOWN));
283                 return -EFSCORRUPTED;
284         }
285         if (xfs_sb_has_incompat_log_feature(sbp,
286                         XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
287                 xfs_warn(mp,
288 "Corruption detected in superblock incompatible log features (0x%x)!",
289                         (sbp->sb_features_log_incompat &
290                                         XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
291                 return -EFSCORRUPTED;
292         }
293
294         /*
295          * We can't read verify the sb LSN because the read verifier is called
296          * before the log is allocated and processed. We know the log is set up
297          * before write verifier calls, so check it here.
298          */
299         if (!xfs_log_check_lsn(mp, sbp->sb_lsn))
300                 return -EFSCORRUPTED;
301
302         return 0;
303 }
304
305 /* Check the validity of the SB. */
306 STATIC int
307 xfs_validate_sb_common(
308         struct xfs_mount        *mp,
309         struct xfs_buf          *bp,
310         struct xfs_sb           *sbp)
311 {
312         struct xfs_dsb          *dsb = bp->b_addr;
313         uint32_t                agcount = 0;
314         uint32_t                rem;
315         bool                    has_dalign;
316
317         if (!xfs_verify_magic(bp, dsb->sb_magicnum)) {
318                 xfs_warn(mp,
319 "Superblock has bad magic number 0x%x. Not an XFS filesystem?",
320                         be32_to_cpu(dsb->sb_magicnum));
321                 return -EWRONGFS;
322         }
323
324         if (!xfs_sb_good_version(sbp)) {
325                 xfs_warn(mp,
326 "Superblock has unknown features enabled or corrupted feature masks.");
327                 return -EWRONGFS;
328         }
329
330         /*
331          * Validate feature flags and state
332          */
333         if (xfs_sb_is_v5(sbp)) {
334                 if (sbp->sb_blocksize < XFS_MIN_CRC_BLOCKSIZE) {
335                         xfs_notice(mp,
336 "Block size (%u bytes) too small for Version 5 superblock (minimum %d bytes)",
337                                 sbp->sb_blocksize, XFS_MIN_CRC_BLOCKSIZE);
338                         return -EFSCORRUPTED;
339                 }
340
341                 /* V5 has a separate project quota inode */
342                 if (sbp->sb_qflags & (XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD)) {
343                         xfs_notice(mp,
344                            "Version 5 of Super block has XFS_OQUOTA bits.");
345                         return -EFSCORRUPTED;
346                 }
347
348                 /*
349                  * Full inode chunks must be aligned to inode chunk size when
350                  * sparse inodes are enabled to support the sparse chunk
351                  * allocation algorithm and prevent overlapping inode records.
352                  */
353                 if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_SPINODES) {
354                         uint32_t        align;
355
356                         align = XFS_INODES_PER_CHUNK * sbp->sb_inodesize
357                                         >> sbp->sb_blocklog;
358                         if (sbp->sb_inoalignmt != align) {
359                                 xfs_warn(mp,
360 "Inode block alignment (%u) must match chunk size (%u) for sparse inodes.",
361                                          sbp->sb_inoalignmt, align);
362                                 return -EINVAL;
363                         }
364                 }
365         } else if (sbp->sb_qflags & (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD |
366                                 XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD)) {
367                         xfs_notice(mp,
368 "Superblock earlier than Version 5 has XFS_{P|G}QUOTA_{ENFD|CHKD} bits.");
369                         return -EFSCORRUPTED;
370         }
371
372         if (unlikely(
373             sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
374                 xfs_warn(mp,
375                 "filesystem is marked as having an external log; "
376                 "specify logdev on the mount command line.");
377                 return -EINVAL;
378         }
379
380         if (unlikely(
381             sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
382                 xfs_warn(mp,
383                 "filesystem is marked as having an internal log; "
384                 "do not specify logdev on the mount command line.");
385                 return -EINVAL;
386         }
387
388         /* Compute agcount for this number of dblocks and agblocks */
389         if (sbp->sb_agblocks) {
390                 agcount = div_u64_rem(sbp->sb_dblocks, sbp->sb_agblocks, &rem);
391                 if (rem)
392                         agcount++;
393         }
394
395         /*
396          * More sanity checking.  Most of these were stolen directly from
397          * xfs_repair.
398          */
399         if (unlikely(
400             sbp->sb_agcount <= 0                                        ||
401             sbp->sb_sectsize < XFS_MIN_SECTORSIZE                       ||
402             sbp->sb_sectsize > XFS_MAX_SECTORSIZE                       ||
403             sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG                    ||
404             sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG                    ||
405             sbp->sb_sectsize != (1 << sbp->sb_sectlog)                  ||
406             sbp->sb_blocksize < XFS_MIN_BLOCKSIZE                       ||
407             sbp->sb_blocksize > XFS_MAX_BLOCKSIZE                       ||
408             sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG                    ||
409             sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG                    ||
410             sbp->sb_blocksize != (1 << sbp->sb_blocklog)                ||
411             sbp->sb_dirblklog + sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
412             sbp->sb_inodesize < XFS_DINODE_MIN_SIZE                     ||
413             sbp->sb_inodesize > XFS_DINODE_MAX_SIZE                     ||
414             sbp->sb_inodelog < XFS_DINODE_MIN_LOG                       ||
415             sbp->sb_inodelog > XFS_DINODE_MAX_LOG                       ||
416             sbp->sb_inodesize != (1 << sbp->sb_inodelog)                ||
417             sbp->sb_inopblock != howmany(sbp->sb_blocksize,sbp->sb_inodesize) ||
418             XFS_FSB_TO_B(mp, sbp->sb_agblocks) < XFS_MIN_AG_BYTES       ||
419             XFS_FSB_TO_B(mp, sbp->sb_agblocks) > XFS_MAX_AG_BYTES       ||
420             sbp->sb_agblklog != xfs_highbit32(sbp->sb_agblocks - 1) + 1 ||
421             agcount == 0 || agcount != sbp->sb_agcount                  ||
422             (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog)   ||
423             (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE)  ||
424             (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE)  ||
425             (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */)    ||
426             sbp->sb_dblocks == 0                                        ||
427             sbp->sb_dblocks > XFS_MAX_DBLOCKS(sbp)                      ||
428             sbp->sb_dblocks < XFS_MIN_DBLOCKS(sbp)                      ||
429             sbp->sb_shared_vn != 0)) {
430                 xfs_notice(mp, "SB sanity check failed");
431                 return -EFSCORRUPTED;
432         }
433
434         /*
435          * Logs that are too large are not supported at all. Reject them
436          * outright. Logs that are too small are tolerated on v4 filesystems,
437          * but we can only check that when mounting the log. Hence we skip
438          * those checks here.
439          */
440         if (sbp->sb_logblocks > XFS_MAX_LOG_BLOCKS) {
441                 xfs_notice(mp,
442                 "Log size 0x%x blocks too large, maximum size is 0x%llx blocks",
443                          sbp->sb_logblocks, XFS_MAX_LOG_BLOCKS);
444                 return -EFSCORRUPTED;
445         }
446
447         if (XFS_FSB_TO_B(mp, sbp->sb_logblocks) > XFS_MAX_LOG_BYTES) {
448                 xfs_warn(mp,
449                 "log size 0x%llx bytes too large, maximum size is 0x%llx bytes",
450                          XFS_FSB_TO_B(mp, sbp->sb_logblocks),
451                          XFS_MAX_LOG_BYTES);
452                 return -EFSCORRUPTED;
453         }
454
455         /*
456          * Do not allow filesystems with corrupted log sector or stripe units to
457          * be mounted. We cannot safely size the iclogs or write to the log if
458          * the log stripe unit is not valid.
459          */
460         if (sbp->sb_versionnum & XFS_SB_VERSION_SECTORBIT) {
461                 if (sbp->sb_logsectsize != (1U << sbp->sb_logsectlog)) {
462                         xfs_notice(mp,
463                         "log sector size in bytes/log2 (0x%x/0x%x) must match",
464                                 sbp->sb_logsectsize, 1U << sbp->sb_logsectlog);
465                         return -EFSCORRUPTED;
466                 }
467         } else if (sbp->sb_logsectsize || sbp->sb_logsectlog) {
468                 xfs_notice(mp,
469                 "log sector size in bytes/log2 (0x%x/0x%x) are not zero",
470                         sbp->sb_logsectsize, sbp->sb_logsectlog);
471                 return -EFSCORRUPTED;
472         }
473
474         if (sbp->sb_logsunit > 1) {
475                 if (sbp->sb_logsunit % sbp->sb_blocksize) {
476                         xfs_notice(mp,
477                 "log stripe unit 0x%x bytes must be a multiple of block size",
478                                 sbp->sb_logsunit);
479                         return -EFSCORRUPTED;
480                 }
481                 if (sbp->sb_logsunit > XLOG_MAX_RECORD_BSIZE) {
482                         xfs_notice(mp,
483                 "log stripe unit 0x%x bytes over maximum size (0x%x bytes)",
484                                 sbp->sb_logsunit, XLOG_MAX_RECORD_BSIZE);
485                         return -EFSCORRUPTED;
486                 }
487         }
488
489         /* Validate the realtime geometry; stolen from xfs_repair */
490         if (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE ||
491             sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) {
492                 xfs_notice(mp,
493                         "realtime extent sanity check failed");
494                 return -EFSCORRUPTED;
495         }
496
497         if (sbp->sb_rblocks == 0) {
498                 if (sbp->sb_rextents != 0 || sbp->sb_rbmblocks != 0 ||
499                     sbp->sb_rextslog != 0 || sbp->sb_frextents != 0) {
500                         xfs_notice(mp,
501                                 "realtime zeroed geometry check failed");
502                         return -EFSCORRUPTED;
503                 }
504         } else {
505                 uint64_t        rexts;
506                 uint64_t        rbmblocks;
507
508                 rexts = div_u64(sbp->sb_rblocks, sbp->sb_rextsize);
509                 rbmblocks = howmany_64(sbp->sb_rextents,
510                                        NBBY * sbp->sb_blocksize);
511
512                 if (!xfs_validate_rtextents(rexts) ||
513                     sbp->sb_rextents != rexts ||
514                     sbp->sb_rextslog != xfs_compute_rextslog(rexts) ||
515                     sbp->sb_rbmblocks != rbmblocks) {
516                         xfs_notice(mp,
517                                 "realtime geometry sanity check failed");
518                         return -EFSCORRUPTED;
519                 }
520         }
521
522         /*
523          * Either (sb_unit and !hasdalign) or (!sb_unit and hasdalign)
524          * would imply the image is corrupted.
525          */
526         has_dalign = sbp->sb_versionnum & XFS_SB_VERSION_DALIGNBIT;
527         if (!!sbp->sb_unit ^ has_dalign) {
528                 xfs_notice(mp, "SB stripe alignment sanity check failed");
529                 return -EFSCORRUPTED;
530         }
531
532         if (!xfs_validate_stripe_geometry(mp, XFS_FSB_TO_B(mp, sbp->sb_unit),
533                         XFS_FSB_TO_B(mp, sbp->sb_width), 0, false))
534                 return -EFSCORRUPTED;
535
536         /*
537          * Currently only very few inode sizes are supported.
538          */
539         switch (sbp->sb_inodesize) {
540         case 256:
541         case 512:
542         case 1024:
543         case 2048:
544                 break;
545         default:
546                 xfs_warn(mp, "inode size of %d bytes not supported",
547                                 sbp->sb_inodesize);
548                 return -ENOSYS;
549         }
550
551         return 0;
552 }
553
554 void
555 xfs_sb_quota_from_disk(struct xfs_sb *sbp)
556 {
557         /*
558          * older mkfs doesn't initialize quota inodes to NULLFSINO. This
559          * leads to in-core values having two different values for a quota
560          * inode to be invalid: 0 and NULLFSINO. Change it to a single value
561          * NULLFSINO.
562          *
563          * Note that this change affect only the in-core values. These
564          * values are not written back to disk unless any quota information
565          * is written to the disk. Even in that case, sb_pquotino field is
566          * not written to disk unless the superblock supports pquotino.
567          */
568         if (sbp->sb_uquotino == 0)
569                 sbp->sb_uquotino = NULLFSINO;
570         if (sbp->sb_gquotino == 0)
571                 sbp->sb_gquotino = NULLFSINO;
572         if (sbp->sb_pquotino == 0)
573                 sbp->sb_pquotino = NULLFSINO;
574
575         /*
576          * We need to do these manipilations only if we are working
577          * with an older version of on-disk superblock.
578          */
579         if (xfs_sb_is_v5(sbp))
580                 return;
581
582         if (sbp->sb_qflags & XFS_OQUOTA_ENFD)
583                 sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
584                                         XFS_PQUOTA_ENFD : XFS_GQUOTA_ENFD;
585         if (sbp->sb_qflags & XFS_OQUOTA_CHKD)
586                 sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
587                                         XFS_PQUOTA_CHKD : XFS_GQUOTA_CHKD;
588         sbp->sb_qflags &= ~(XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD);
589
590         if (sbp->sb_qflags & XFS_PQUOTA_ACCT &&
591             sbp->sb_gquotino != NULLFSINO)  {
592                 /*
593                  * In older version of superblock, on-disk superblock only
594                  * has sb_gquotino, and in-core superblock has both sb_gquotino
595                  * and sb_pquotino. But, only one of them is supported at any
596                  * point of time. So, if PQUOTA is set in disk superblock,
597                  * copy over sb_gquotino to sb_pquotino.  The NULLFSINO test
598                  * above is to make sure we don't do this twice and wipe them
599                  * both out!
600                  */
601                 sbp->sb_pquotino = sbp->sb_gquotino;
602                 sbp->sb_gquotino = NULLFSINO;
603         }
604 }
605
606 static void
607 __xfs_sb_from_disk(
608         struct xfs_sb   *to,
609         struct xfs_dsb  *from,
610         bool            convert_xquota)
611 {
612         to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
613         to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
614         to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
615         to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
616         to->sb_rextents = be64_to_cpu(from->sb_rextents);
617         memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
618         to->sb_logstart = be64_to_cpu(from->sb_logstart);
619         to->sb_rootino = be64_to_cpu(from->sb_rootino);
620         to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
621         to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
622         to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
623         to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
624         to->sb_agcount = be32_to_cpu(from->sb_agcount);
625         to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
626         to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
627         to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
628         to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
629         to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
630         to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
631         memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
632         to->sb_blocklog = from->sb_blocklog;
633         to->sb_sectlog = from->sb_sectlog;
634         to->sb_inodelog = from->sb_inodelog;
635         to->sb_inopblog = from->sb_inopblog;
636         to->sb_agblklog = from->sb_agblklog;
637         to->sb_rextslog = from->sb_rextslog;
638         to->sb_inprogress = from->sb_inprogress;
639         to->sb_imax_pct = from->sb_imax_pct;
640         to->sb_icount = be64_to_cpu(from->sb_icount);
641         to->sb_ifree = be64_to_cpu(from->sb_ifree);
642         to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
643         to->sb_frextents = be64_to_cpu(from->sb_frextents);
644         to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
645         to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
646         to->sb_qflags = be16_to_cpu(from->sb_qflags);
647         to->sb_flags = from->sb_flags;
648         to->sb_shared_vn = from->sb_shared_vn;
649         to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
650         to->sb_unit = be32_to_cpu(from->sb_unit);
651         to->sb_width = be32_to_cpu(from->sb_width);
652         to->sb_dirblklog = from->sb_dirblklog;
653         to->sb_logsectlog = from->sb_logsectlog;
654         to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
655         to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
656         to->sb_features2 = be32_to_cpu(from->sb_features2);
657         to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
658         to->sb_features_compat = be32_to_cpu(from->sb_features_compat);
659         to->sb_features_ro_compat = be32_to_cpu(from->sb_features_ro_compat);
660         to->sb_features_incompat = be32_to_cpu(from->sb_features_incompat);
661         to->sb_features_log_incompat =
662                                 be32_to_cpu(from->sb_features_log_incompat);
663         /* crc is only used on disk, not in memory; just init to 0 here. */
664         to->sb_crc = 0;
665         to->sb_spino_align = be32_to_cpu(from->sb_spino_align);
666         to->sb_pquotino = be64_to_cpu(from->sb_pquotino);
667         to->sb_lsn = be64_to_cpu(from->sb_lsn);
668         /*
669          * sb_meta_uuid is only on disk if it differs from sb_uuid and the
670          * feature flag is set; if not set we keep it only in memory.
671          */
672         if (xfs_sb_is_v5(to) &&
673             (to->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_META_UUID))
674                 uuid_copy(&to->sb_meta_uuid, &from->sb_meta_uuid);
675         else
676                 uuid_copy(&to->sb_meta_uuid, &from->sb_uuid);
677         /* Convert on-disk flags to in-memory flags? */
678         if (convert_xquota)
679                 xfs_sb_quota_from_disk(to);
680 }
681
682 void
683 xfs_sb_from_disk(
684         struct xfs_sb   *to,
685         struct xfs_dsb  *from)
686 {
687         __xfs_sb_from_disk(to, from, true);
688 }
689
690 static void
691 xfs_sb_quota_to_disk(
692         struct xfs_dsb  *to,
693         struct xfs_sb   *from)
694 {
695         uint16_t        qflags = from->sb_qflags;
696
697         to->sb_uquotino = cpu_to_be64(from->sb_uquotino);
698
699         /*
700          * The in-memory superblock quota state matches the v5 on-disk format so
701          * just write them out and return
702          */
703         if (xfs_sb_is_v5(from)) {
704                 to->sb_qflags = cpu_to_be16(from->sb_qflags);
705                 to->sb_gquotino = cpu_to_be64(from->sb_gquotino);
706                 to->sb_pquotino = cpu_to_be64(from->sb_pquotino);
707                 return;
708         }
709
710         /*
711          * For older superblocks (v4), the in-core version of sb_qflags do not
712          * have XFS_OQUOTA_* flags, whereas the on-disk version does.  So,
713          * convert incore XFS_{PG}QUOTA_* flags to on-disk XFS_OQUOTA_* flags.
714          */
715         qflags &= ~(XFS_PQUOTA_ENFD | XFS_PQUOTA_CHKD |
716                         XFS_GQUOTA_ENFD | XFS_GQUOTA_CHKD);
717
718         if (from->sb_qflags &
719                         (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD))
720                 qflags |= XFS_OQUOTA_ENFD;
721         if (from->sb_qflags &
722                         (XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD))
723                 qflags |= XFS_OQUOTA_CHKD;
724         to->sb_qflags = cpu_to_be16(qflags);
725
726         /*
727          * GQUOTINO and PQUOTINO cannot be used together in versions
728          * of superblock that do not have pquotino. from->sb_flags
729          * tells us which quota is active and should be copied to
730          * disk. If neither are active, we should NULL the inode.
731          *
732          * In all cases, the separate pquotino must remain 0 because it
733          * is beyond the "end" of the valid non-pquotino superblock.
734          */
735         if (from->sb_qflags & XFS_GQUOTA_ACCT)
736                 to->sb_gquotino = cpu_to_be64(from->sb_gquotino);
737         else if (from->sb_qflags & XFS_PQUOTA_ACCT)
738                 to->sb_gquotino = cpu_to_be64(from->sb_pquotino);
739         else {
740                 /*
741                  * We can't rely on just the fields being logged to tell us
742                  * that it is safe to write NULLFSINO - we should only do that
743                  * if quotas are not actually enabled. Hence only write
744                  * NULLFSINO if both in-core quota inodes are NULL.
745                  */
746                 if (from->sb_gquotino == NULLFSINO &&
747                     from->sb_pquotino == NULLFSINO)
748                         to->sb_gquotino = cpu_to_be64(NULLFSINO);
749         }
750
751         to->sb_pquotino = 0;
752 }
753
754 void
755 xfs_sb_to_disk(
756         struct xfs_dsb  *to,
757         struct xfs_sb   *from)
758 {
759         xfs_sb_quota_to_disk(to, from);
760
761         to->sb_magicnum = cpu_to_be32(from->sb_magicnum);
762         to->sb_blocksize = cpu_to_be32(from->sb_blocksize);
763         to->sb_dblocks = cpu_to_be64(from->sb_dblocks);
764         to->sb_rblocks = cpu_to_be64(from->sb_rblocks);
765         to->sb_rextents = cpu_to_be64(from->sb_rextents);
766         memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
767         to->sb_logstart = cpu_to_be64(from->sb_logstart);
768         to->sb_rootino = cpu_to_be64(from->sb_rootino);
769         to->sb_rbmino = cpu_to_be64(from->sb_rbmino);
770         to->sb_rsumino = cpu_to_be64(from->sb_rsumino);
771         to->sb_rextsize = cpu_to_be32(from->sb_rextsize);
772         to->sb_agblocks = cpu_to_be32(from->sb_agblocks);
773         to->sb_agcount = cpu_to_be32(from->sb_agcount);
774         to->sb_rbmblocks = cpu_to_be32(from->sb_rbmblocks);
775         to->sb_logblocks = cpu_to_be32(from->sb_logblocks);
776         to->sb_versionnum = cpu_to_be16(from->sb_versionnum);
777         to->sb_sectsize = cpu_to_be16(from->sb_sectsize);
778         to->sb_inodesize = cpu_to_be16(from->sb_inodesize);
779         to->sb_inopblock = cpu_to_be16(from->sb_inopblock);
780         memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
781         to->sb_blocklog = from->sb_blocklog;
782         to->sb_sectlog = from->sb_sectlog;
783         to->sb_inodelog = from->sb_inodelog;
784         to->sb_inopblog = from->sb_inopblog;
785         to->sb_agblklog = from->sb_agblklog;
786         to->sb_rextslog = from->sb_rextslog;
787         to->sb_inprogress = from->sb_inprogress;
788         to->sb_imax_pct = from->sb_imax_pct;
789         to->sb_icount = cpu_to_be64(from->sb_icount);
790         to->sb_ifree = cpu_to_be64(from->sb_ifree);
791         to->sb_fdblocks = cpu_to_be64(from->sb_fdblocks);
792         to->sb_frextents = cpu_to_be64(from->sb_frextents);
793
794         to->sb_flags = from->sb_flags;
795         to->sb_shared_vn = from->sb_shared_vn;
796         to->sb_inoalignmt = cpu_to_be32(from->sb_inoalignmt);
797         to->sb_unit = cpu_to_be32(from->sb_unit);
798         to->sb_width = cpu_to_be32(from->sb_width);
799         to->sb_dirblklog = from->sb_dirblklog;
800         to->sb_logsectlog = from->sb_logsectlog;
801         to->sb_logsectsize = cpu_to_be16(from->sb_logsectsize);
802         to->sb_logsunit = cpu_to_be32(from->sb_logsunit);
803
804         /*
805          * We need to ensure that bad_features2 always matches features2.
806          * Hence we enforce that here rather than having to remember to do it
807          * everywhere else that updates features2.
808          */
809         from->sb_bad_features2 = from->sb_features2;
810         to->sb_features2 = cpu_to_be32(from->sb_features2);
811         to->sb_bad_features2 = cpu_to_be32(from->sb_bad_features2);
812
813         if (!xfs_sb_is_v5(from))
814                 return;
815
816         to->sb_features_compat = cpu_to_be32(from->sb_features_compat);
817         to->sb_features_ro_compat =
818                         cpu_to_be32(from->sb_features_ro_compat);
819         to->sb_features_incompat =
820                         cpu_to_be32(from->sb_features_incompat);
821         to->sb_features_log_incompat =
822                         cpu_to_be32(from->sb_features_log_incompat);
823         to->sb_spino_align = cpu_to_be32(from->sb_spino_align);
824         to->sb_lsn = cpu_to_be64(from->sb_lsn);
825         if (from->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_META_UUID)
826                 uuid_copy(&to->sb_meta_uuid, &from->sb_meta_uuid);
827 }
828
829 /*
830  * If the superblock has the CRC feature bit set or the CRC field is non-null,
831  * check that the CRC is valid.  We check the CRC field is non-null because a
832  * single bit error could clear the feature bit and unused parts of the
833  * superblock are supposed to be zero. Hence a non-null crc field indicates that
834  * we've potentially lost a feature bit and we should check it anyway.
835  *
836  * However, past bugs (i.e. in growfs) left non-zeroed regions beyond the
837  * last field in V4 secondary superblocks.  So for secondary superblocks,
838  * we are more forgiving, and ignore CRC failures if the primary doesn't
839  * indicate that the fs version is V5.
840  */
841 static void
842 xfs_sb_read_verify(
843         struct xfs_buf          *bp)
844 {
845         struct xfs_sb           sb;
846         struct xfs_mount        *mp = bp->b_mount;
847         struct xfs_dsb          *dsb = bp->b_addr;
848         int                     error;
849
850         /*
851          * open code the version check to avoid needing to convert the entire
852          * superblock from disk order just to check the version number
853          */
854         if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC) &&
855             (((be16_to_cpu(dsb->sb_versionnum) & XFS_SB_VERSION_NUMBITS) ==
856                                                 XFS_SB_VERSION_5) ||
857              dsb->sb_crc != 0)) {
858
859                 if (!xfs_buf_verify_cksum(bp, XFS_SB_CRC_OFF)) {
860                         /* Only fail bad secondaries on a known V5 filesystem */
861                         if (xfs_buf_daddr(bp) == XFS_SB_DADDR ||
862                             xfs_has_crc(mp)) {
863                                 error = -EFSBADCRC;
864                                 goto out_error;
865                         }
866                 }
867         }
868
869         /*
870          * Check all the superblock fields.  Don't byteswap the xquota flags
871          * because _verify_common checks the on-disk values.
872          */
873         __xfs_sb_from_disk(&sb, dsb, false);
874         error = xfs_validate_sb_common(mp, bp, &sb);
875         if (error)
876                 goto out_error;
877         error = xfs_validate_sb_read(mp, &sb);
878
879 out_error:
880         if (error == -EFSCORRUPTED || error == -EFSBADCRC)
881                 xfs_verifier_error(bp, error, __this_address);
882         else if (error)
883                 xfs_buf_ioerror(bp, error);
884 }
885
886 /*
887  * We may be probed for a filesystem match, so we may not want to emit
888  * messages when the superblock buffer is not actually an XFS superblock.
889  * If we find an XFS superblock, then run a normal, noisy mount because we are
890  * really going to mount it and want to know about errors.
891  */
892 static void
893 xfs_sb_quiet_read_verify(
894         struct xfs_buf  *bp)
895 {
896         struct xfs_dsb  *dsb = bp->b_addr;
897
898         if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC)) {
899                 /* XFS filesystem, verify noisily! */
900                 xfs_sb_read_verify(bp);
901                 return;
902         }
903         /* quietly fail */
904         xfs_buf_ioerror(bp, -EWRONGFS);
905 }
906
907 static void
908 xfs_sb_write_verify(
909         struct xfs_buf          *bp)
910 {
911         struct xfs_sb           sb;
912         struct xfs_mount        *mp = bp->b_mount;
913         struct xfs_buf_log_item *bip = bp->b_log_item;
914         struct xfs_dsb          *dsb = bp->b_addr;
915         int                     error;
916
917         /*
918          * Check all the superblock fields.  Don't byteswap the xquota flags
919          * because _verify_common checks the on-disk values.
920          */
921         __xfs_sb_from_disk(&sb, dsb, false);
922         error = xfs_validate_sb_common(mp, bp, &sb);
923         if (error)
924                 goto out_error;
925         error = xfs_validate_sb_write(mp, bp, &sb);
926         if (error)
927                 goto out_error;
928
929         if (!xfs_sb_is_v5(&sb))
930                 return;
931
932         if (bip)
933                 dsb->sb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
934
935         xfs_buf_update_cksum(bp, XFS_SB_CRC_OFF);
936         return;
937
938 out_error:
939         xfs_verifier_error(bp, error, __this_address);
940 }
941
942 const struct xfs_buf_ops xfs_sb_buf_ops = {
943         .name = "xfs_sb",
944         .magic = { cpu_to_be32(XFS_SB_MAGIC), cpu_to_be32(XFS_SB_MAGIC) },
945         .verify_read = xfs_sb_read_verify,
946         .verify_write = xfs_sb_write_verify,
947 };
948
949 const struct xfs_buf_ops xfs_sb_quiet_buf_ops = {
950         .name = "xfs_sb_quiet",
951         .magic = { cpu_to_be32(XFS_SB_MAGIC), cpu_to_be32(XFS_SB_MAGIC) },
952         .verify_read = xfs_sb_quiet_read_verify,
953         .verify_write = xfs_sb_write_verify,
954 };
955
956 /*
957  * xfs_mount_common
958  *
959  * Mount initialization code establishing various mount
960  * fields from the superblock associated with the given
961  * mount structure.
962  *
963  * Inode geometry are calculated in xfs_ialloc_setup_geometry.
964  */
965 void
966 xfs_sb_mount_common(
967         struct xfs_mount        *mp,
968         struct xfs_sb           *sbp)
969 {
970         mp->m_agfrotor = 0;
971         atomic_set(&mp->m_agirotor, 0);
972         mp->m_maxagi = mp->m_sb.sb_agcount;
973         mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
974         mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
975         mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
976         mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
977         mp->m_blockmask = sbp->sb_blocksize - 1;
978         mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
979         mp->m_blockwmask = mp->m_blockwsize - 1;
980         mp->m_rtxblklog = log2_if_power2(sbp->sb_rextsize);
981         mp->m_rtxblkmask = mask64_if_power2(sbp->sb_rextsize);
982
983         mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
984         mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
985         mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
986         mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
987
988         mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
989         mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
990         mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
991         mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
992
993         mp->m_rmap_mxr[0] = xfs_rmapbt_maxrecs(sbp->sb_blocksize, 1);
994         mp->m_rmap_mxr[1] = xfs_rmapbt_maxrecs(sbp->sb_blocksize, 0);
995         mp->m_rmap_mnr[0] = mp->m_rmap_mxr[0] / 2;
996         mp->m_rmap_mnr[1] = mp->m_rmap_mxr[1] / 2;
997
998         mp->m_refc_mxr[0] = xfs_refcountbt_maxrecs(sbp->sb_blocksize, true);
999         mp->m_refc_mxr[1] = xfs_refcountbt_maxrecs(sbp->sb_blocksize, false);
1000         mp->m_refc_mnr[0] = mp->m_refc_mxr[0] / 2;
1001         mp->m_refc_mnr[1] = mp->m_refc_mxr[1] / 2;
1002
1003         mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
1004         mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
1005         mp->m_ag_max_usable = xfs_alloc_ag_max_usable(mp);
1006 }
1007
1008 /*
1009  * xfs_log_sb() can be used to copy arbitrary changes to the in-core superblock
1010  * into the superblock buffer to be logged.  It does not provide the higher
1011  * level of locking that is needed to protect the in-core superblock from
1012  * concurrent access.
1013  */
1014 void
1015 xfs_log_sb(
1016         struct xfs_trans        *tp)
1017 {
1018         struct xfs_mount        *mp = tp->t_mountp;
1019         struct xfs_buf          *bp = xfs_trans_getsb(tp);
1020
1021         /*
1022          * Lazy sb counters don't update the in-core superblock so do that now.
1023          * If this is at unmount, the counters will be exactly correct, but at
1024          * any other time they will only be ballpark correct because of
1025          * reservations that have been taken out percpu counters. If we have an
1026          * unclean shutdown, this will be corrected by log recovery rebuilding
1027          * the counters from the AGF block counts.
1028          *
1029          * Do not update sb_frextents here because it is not part of the lazy
1030          * sb counters, despite having a percpu counter. It is always kept
1031          * consistent with the ondisk rtbitmap by xfs_trans_apply_sb_deltas()
1032          * and hence we don't need have to update it here.
1033          */
1034         if (xfs_has_lazysbcount(mp)) {
1035                 mp->m_sb.sb_icount = percpu_counter_sum(&mp->m_icount);
1036                 mp->m_sb.sb_ifree = min_t(uint64_t,
1037                                 percpu_counter_sum(&mp->m_ifree),
1038                                 mp->m_sb.sb_icount);
1039                 mp->m_sb.sb_fdblocks = percpu_counter_sum(&mp->m_fdblocks);
1040         }
1041
1042         xfs_sb_to_disk(bp->b_addr, &mp->m_sb);
1043         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
1044         xfs_trans_log_buf(tp, bp, 0, sizeof(struct xfs_dsb) - 1);
1045 }
1046
1047 /*
1048  * xfs_sync_sb
1049  *
1050  * Sync the superblock to disk.
1051  *
1052  * Note that the caller is responsible for checking the frozen state of the
1053  * filesystem. This procedure uses the non-blocking transaction allocator and
1054  * thus will allow modifications to a frozen fs. This is required because this
1055  * code can be called during the process of freezing where use of the high-level
1056  * allocator would deadlock.
1057  */
1058 int
1059 xfs_sync_sb(
1060         struct xfs_mount        *mp,
1061         bool                    wait)
1062 {
1063         struct xfs_trans        *tp;
1064         int                     error;
1065
1066         error = xfs_trans_alloc(mp, &M_RES(mp)->tr_sb, 0, 0,
1067                         XFS_TRANS_NO_WRITECOUNT, &tp);
1068         if (error)
1069                 return error;
1070
1071         xfs_log_sb(tp);
1072         if (wait)
1073                 xfs_trans_set_sync(tp);
1074         return xfs_trans_commit(tp);
1075 }
1076
1077 /*
1078  * Update all the secondary superblocks to match the new state of the primary.
1079  * Because we are completely overwriting all the existing fields in the
1080  * secondary superblock buffers, there is no need to read them in from disk.
1081  * Just get a new buffer, stamp it and write it.
1082  *
1083  * The sb buffers need to be cached here so that we serialise against other
1084  * operations that access the secondary superblocks, but we don't want to keep
1085  * them in memory once it is written so we mark it as a one-shot buffer.
1086  */
1087 int
1088 xfs_update_secondary_sbs(
1089         struct xfs_mount        *mp)
1090 {
1091         struct xfs_perag        *pag;
1092         xfs_agnumber_t          agno = 1;
1093         int                     saved_error = 0;
1094         int                     error = 0;
1095         LIST_HEAD               (buffer_list);
1096
1097         /* update secondary superblocks. */
1098         for_each_perag_from(mp, agno, pag) {
1099                 struct xfs_buf          *bp;
1100
1101                 error = xfs_buf_get(mp->m_ddev_targp,
1102                                  XFS_AG_DADDR(mp, pag->pag_agno, XFS_SB_DADDR),
1103                                  XFS_FSS_TO_BB(mp, 1), &bp);
1104                 /*
1105                  * If we get an error reading or writing alternate superblocks,
1106                  * continue.  xfs_repair chooses the "best" superblock based
1107                  * on most matches; if we break early, we'll leave more
1108                  * superblocks un-updated than updated, and xfs_repair may
1109                  * pick them over the properly-updated primary.
1110                  */
1111                 if (error) {
1112                         xfs_warn(mp,
1113                 "error allocating secondary superblock for ag %d",
1114                                 pag->pag_agno);
1115                         if (!saved_error)
1116                                 saved_error = error;
1117                         continue;
1118                 }
1119
1120                 bp->b_ops = &xfs_sb_buf_ops;
1121                 xfs_buf_oneshot(bp);
1122                 xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
1123                 xfs_sb_to_disk(bp->b_addr, &mp->m_sb);
1124                 xfs_buf_delwri_queue(bp, &buffer_list);
1125                 xfs_buf_relse(bp);
1126
1127                 /* don't hold too many buffers at once */
1128                 if (agno % 16)
1129                         continue;
1130
1131                 error = xfs_buf_delwri_submit(&buffer_list);
1132                 if (error) {
1133                         xfs_warn(mp,
1134                 "write error %d updating a secondary superblock near ag %d",
1135                                 error, pag->pag_agno);
1136                         if (!saved_error)
1137                                 saved_error = error;
1138                         continue;
1139                 }
1140         }
1141         error = xfs_buf_delwri_submit(&buffer_list);
1142         if (error) {
1143                 xfs_warn(mp,
1144                 "write error %d updating a secondary superblock near ag %d",
1145                         error, agno);
1146         }
1147
1148         return saved_error ? saved_error : error;
1149 }
1150
1151 /*
1152  * Same behavior as xfs_sync_sb, except that it is always synchronous and it
1153  * also writes the superblock buffer to disk sector 0 immediately.
1154  */
1155 int
1156 xfs_sync_sb_buf(
1157         struct xfs_mount        *mp)
1158 {
1159         struct xfs_trans        *tp;
1160         struct xfs_buf          *bp;
1161         int                     error;
1162
1163         error = xfs_trans_alloc(mp, &M_RES(mp)->tr_sb, 0, 0, 0, &tp);
1164         if (error)
1165                 return error;
1166
1167         bp = xfs_trans_getsb(tp);
1168         xfs_log_sb(tp);
1169         xfs_trans_bhold(tp, bp);
1170         xfs_trans_set_sync(tp);
1171         error = xfs_trans_commit(tp);
1172         if (error)
1173                 goto out;
1174         /*
1175          * write out the sb buffer to get the changes to disk
1176          */
1177         error = xfs_bwrite(bp);
1178 out:
1179         xfs_buf_relse(bp);
1180         return error;
1181 }
1182
1183 void
1184 xfs_fs_geometry(
1185         struct xfs_mount        *mp,
1186         struct xfs_fsop_geom    *geo,
1187         int                     struct_version)
1188 {
1189         struct xfs_sb           *sbp = &mp->m_sb;
1190
1191         memset(geo, 0, sizeof(struct xfs_fsop_geom));
1192
1193         geo->blocksize = sbp->sb_blocksize;
1194         geo->rtextsize = sbp->sb_rextsize;
1195         geo->agblocks = sbp->sb_agblocks;
1196         geo->agcount = sbp->sb_agcount;
1197         geo->logblocks = sbp->sb_logblocks;
1198         geo->sectsize = sbp->sb_sectsize;
1199         geo->inodesize = sbp->sb_inodesize;
1200         geo->imaxpct = sbp->sb_imax_pct;
1201         geo->datablocks = sbp->sb_dblocks;
1202         geo->rtblocks = sbp->sb_rblocks;
1203         geo->rtextents = sbp->sb_rextents;
1204         geo->logstart = sbp->sb_logstart;
1205         BUILD_BUG_ON(sizeof(geo->uuid) != sizeof(sbp->sb_uuid));
1206         memcpy(geo->uuid, &sbp->sb_uuid, sizeof(sbp->sb_uuid));
1207
1208         if (struct_version < 2)
1209                 return;
1210
1211         geo->sunit = sbp->sb_unit;
1212         geo->swidth = sbp->sb_width;
1213
1214         if (struct_version < 3)
1215                 return;
1216
1217         geo->version = XFS_FSOP_GEOM_VERSION;
1218         geo->flags = XFS_FSOP_GEOM_FLAGS_NLINK |
1219                      XFS_FSOP_GEOM_FLAGS_DIRV2 |
1220                      XFS_FSOP_GEOM_FLAGS_EXTFLG;
1221         if (xfs_has_attr(mp))
1222                 geo->flags |= XFS_FSOP_GEOM_FLAGS_ATTR;
1223         if (xfs_has_quota(mp))
1224                 geo->flags |= XFS_FSOP_GEOM_FLAGS_QUOTA;
1225         if (xfs_has_align(mp))
1226                 geo->flags |= XFS_FSOP_GEOM_FLAGS_IALIGN;
1227         if (xfs_has_dalign(mp))
1228                 geo->flags |= XFS_FSOP_GEOM_FLAGS_DALIGN;
1229         if (xfs_has_asciici(mp))
1230                 geo->flags |= XFS_FSOP_GEOM_FLAGS_DIRV2CI;
1231         if (xfs_has_lazysbcount(mp))
1232                 geo->flags |= XFS_FSOP_GEOM_FLAGS_LAZYSB;
1233         if (xfs_has_attr2(mp))
1234                 geo->flags |= XFS_FSOP_GEOM_FLAGS_ATTR2;
1235         if (xfs_has_projid32(mp))
1236                 geo->flags |= XFS_FSOP_GEOM_FLAGS_PROJID32;
1237         if (xfs_has_crc(mp))
1238                 geo->flags |= XFS_FSOP_GEOM_FLAGS_V5SB;
1239         if (xfs_has_ftype(mp))
1240                 geo->flags |= XFS_FSOP_GEOM_FLAGS_FTYPE;
1241         if (xfs_has_finobt(mp))
1242                 geo->flags |= XFS_FSOP_GEOM_FLAGS_FINOBT;
1243         if (xfs_has_sparseinodes(mp))
1244                 geo->flags |= XFS_FSOP_GEOM_FLAGS_SPINODES;
1245         if (xfs_has_rmapbt(mp))
1246                 geo->flags |= XFS_FSOP_GEOM_FLAGS_RMAPBT;
1247         if (xfs_has_reflink(mp))
1248                 geo->flags |= XFS_FSOP_GEOM_FLAGS_REFLINK;
1249         if (xfs_has_bigtime(mp))
1250                 geo->flags |= XFS_FSOP_GEOM_FLAGS_BIGTIME;
1251         if (xfs_has_inobtcounts(mp))
1252                 geo->flags |= XFS_FSOP_GEOM_FLAGS_INOBTCNT;
1253         if (xfs_has_sector(mp)) {
1254                 geo->flags |= XFS_FSOP_GEOM_FLAGS_SECTOR;
1255                 geo->logsectsize = sbp->sb_logsectsize;
1256         } else {
1257                 geo->logsectsize = BBSIZE;
1258         }
1259         if (xfs_has_large_extent_counts(mp))
1260                 geo->flags |= XFS_FSOP_GEOM_FLAGS_NREXT64;
1261         geo->rtsectsize = sbp->sb_blocksize;
1262         geo->dirblocksize = xfs_dir2_dirblock_bytes(sbp);
1263
1264         if (struct_version < 4)
1265                 return;
1266
1267         if (xfs_has_logv2(mp))
1268                 geo->flags |= XFS_FSOP_GEOM_FLAGS_LOGV2;
1269
1270         geo->logsunit = sbp->sb_logsunit;
1271
1272         if (struct_version < 5)
1273                 return;
1274
1275         geo->version = XFS_FSOP_GEOM_VERSION_V5;
1276 }
1277
1278 /* Read a secondary superblock. */
1279 int
1280 xfs_sb_read_secondary(
1281         struct xfs_mount        *mp,
1282         struct xfs_trans        *tp,
1283         xfs_agnumber_t          agno,
1284         struct xfs_buf          **bpp)
1285 {
1286         struct xfs_buf          *bp;
1287         int                     error;
1288
1289         ASSERT(agno != 0 && agno != NULLAGNUMBER);
1290         error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
1291                         XFS_AG_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
1292                         XFS_FSS_TO_BB(mp, 1), 0, &bp, &xfs_sb_buf_ops);
1293         if (error)
1294                 return error;
1295         xfs_buf_set_ref(bp, XFS_SSB_REF);
1296         *bpp = bp;
1297         return 0;
1298 }
1299
1300 /* Get an uninitialised secondary superblock buffer. */
1301 int
1302 xfs_sb_get_secondary(
1303         struct xfs_mount        *mp,
1304         struct xfs_trans        *tp,
1305         xfs_agnumber_t          agno,
1306         struct xfs_buf          **bpp)
1307 {
1308         struct xfs_buf          *bp;
1309         int                     error;
1310
1311         ASSERT(agno != 0 && agno != NULLAGNUMBER);
1312         error = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1313                         XFS_AG_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
1314                         XFS_FSS_TO_BB(mp, 1), 0, &bp);
1315         if (error)
1316                 return error;
1317         bp->b_ops = &xfs_sb_buf_ops;
1318         xfs_buf_oneshot(bp);
1319         *bpp = bp;
1320         return 0;
1321 }
1322
1323 /*
1324  * sunit, swidth, sectorsize(optional with 0) should be all in bytes,
1325  * so users won't be confused by values in error messages.
1326  */
1327 bool
1328 xfs_validate_stripe_geometry(
1329         struct xfs_mount        *mp,
1330         __s64                   sunit,
1331         __s64                   swidth,
1332         int                     sectorsize,
1333         bool                    silent)
1334 {
1335         if (swidth > INT_MAX) {
1336                 if (!silent)
1337                         xfs_notice(mp,
1338 "stripe width (%lld) is too large", swidth);
1339                 return false;
1340         }
1341
1342         if (sunit > swidth) {
1343                 if (!silent)
1344                         xfs_notice(mp,
1345 "stripe unit (%lld) is larger than the stripe width (%lld)", sunit, swidth);
1346                 return false;
1347         }
1348
1349         if (sectorsize && (int)sunit % sectorsize) {
1350                 if (!silent)
1351                         xfs_notice(mp,
1352 "stripe unit (%lld) must be a multiple of the sector size (%d)",
1353                                    sunit, sectorsize);
1354                 return false;
1355         }
1356
1357         if (sunit && !swidth) {
1358                 if (!silent)
1359                         xfs_notice(mp,
1360 "invalid stripe unit (%lld) and stripe width of 0", sunit);
1361                 return false;
1362         }
1363
1364         if (!sunit && swidth) {
1365                 if (!silent)
1366                         xfs_notice(mp,
1367 "invalid stripe width (%lld) and stripe unit of 0", swidth);
1368                 return false;
1369         }
1370
1371         if (sunit && (int)swidth % (int)sunit) {
1372                 if (!silent)
1373                         xfs_notice(mp,
1374 "stripe width (%lld) must be a multiple of the stripe unit (%lld)",
1375                                    swidth, sunit);
1376                 return false;
1377         }
1378         return true;
1379 }
1380
1381 /*
1382  * Compute the maximum level number of the realtime summary file, as defined by
1383  * mkfs.  The historic use of highbit32 on a 64-bit quantity prohibited correct
1384  * use of rt volumes with more than 2^32 extents.
1385  */
1386 uint8_t
1387 xfs_compute_rextslog(
1388         xfs_rtbxlen_t           rtextents)
1389 {
1390         if (!rtextents)
1391                 return 0;
1392         return xfs_highbit64(rtextents);
1393 }